Can statistical and entropy-based features extracted from ECG signals efficiently differentiate the cannabis-consuming women population from the non-consumer?

Abstract

Cannabis as bhang is one of the most widely used recreational products in many parts of rural India. Many women who consume cannabis take it secretly due to societal fear. Consumption of cannabis may result in several short-term and long-term health conditions. Hence, detecting the cannabis-consuming women population is needed to avoid its associated health conditions early. The heart rate variability features have been earlier used for analyzing cardiac health; however, the high computational complexity of the method is a disadvantage. The current study hypothesizes that the statistical and entropy features obtained from ECG segments can efficiently differentiate the cannabis-consuming population from the non-consumer in women population. For this purpose, significantly different features were employed as input to various machine learning (ML) models, and the performance of the ML models was evaluated. The results showed that the Gradient Boosted Tree (GBT) model provided the optimal performance (mean accuracy: 75.9%) among the ML models used. The statistically significant features and the higher classification accuracy of the GBT ML model support our presumed hypothesis.